Presteniko Septi R retweeted
We’re introducing Hydrogen-1.0, our first open-weight model for scientific research. We post-trained GLM-5.3 (@Zai_org) on our internal scientific data, lab records, and discovery trajectories.
It scores 38.3% on FrontierScience-Research vs GPT-6-Astra’s 33.3% and scores 93.0% on GPQA-Diamonds.
We believe science should be open, collaborative, and democratized, rather than proprietary. Only open weights let researchers inspect, adapt and build on the unprecedented speed of AI growth. There are many recent models named by chemical elements, but we believe Hydrogen-1.0 is the first aiming towards and open scientific superintelligence for discovery. 🧵
Presteniko Septi R retweeted
そういえば、先月でstable株式会社を創業して、丸3年が経ちました!🎊
この3年間は苦難の連続で、なかなか思うようにいかないことも多くありましたが、関わってくれたあらゆる方のおかげで何とかやってこれています。
最近は、営業と採用をじわじわと強化しており、今後はどんどん面白いことにチャレンジしていきたいと思っています。データエンジニアリングについての課題は、世の中にはまだまだたくさんあるので、我々が出せるバリューを少しでも大きくできたらいいなと思っています!
改めてよろしくお願いします!🙌
Presteniko Septi R retweeted
Kurashiru AI OSから「デジタルツインAgent」をリリースしました。実際の購買データで更新し続ける消費者アバターに、発売前の企画を壁打ちできます。
発売の前に、市場と対話する。マーケティングリサーチは、高くて、遅くて、一度きりですがそこに革新を起こします。安く、24時間、何度でもリサーチ出来ます。
これは本当に使えそうな。特に非構造化データに。試しみて行きたいな。
RAG is officially cooked.
IBM built a retriever that hallucinates 65x less than fine-tuned RAG systems.
Right now, every company is using Retrieval-Augmented Generation (RAG) to stop their AI from hallucinating.
The idea is to give the AI a private database to read so it doesn't make things up.
But there is a massive structural flaw.
Standard RAG blindly grabs documents based on vector similarity and forces the AI to read them. If the retriever pulls irrelevant garbage, the AI doesn't stop.
It just hallucinates with citations.
IBM just dropped a paper that changes the math entirely.
They built a new retrieval architecture that hallucinates 65x less than state-of-the-art fine-tuned RAG systems.
65x is the difference between a fun prototype and a production-grade enterprise system.
Instead of blindly dumping context into the model and hoping for the best, their system actively filters and validates the retrieved data before the generation step even begins.
It cuts out the noise. It forces absolute precision.
For businesses, this is the holy grail.
Zero hallucinations means zero liability. It means autonomous agents can finally be trusted with sensitive company data without a human babysitting every output.
22
I believe the key to build the consumer digital twin if we really want to create a behavioral copy is to do some real interviews. That way we could reduce positivity bias that often came from the surveys.
13
Presteniko Septi R retweeted
Kurashiru AI OSで、レシピ検索閲覧データとレシート購買データから消費者ごとのAIアバターを構築し、新商品コンセプトテストなどを即時に実行できるサービスをリリースしました!
クラシル独自のデータを活用したAIソリューションも今後提供していきます。
クラシル株式会社、生活者の購買・行動傾向を統計データから再現したAIアバターに"何度でも"聞ける「デジタルツインAgent」をKurashiru AI OSで提供開始 prtimes.jp/main/html/rd/p/00…
Presteniko Septi R retweeted
案件拡大に伴い、正社員・業務委託ともに、メンバーを募集しています!ご興味ある方、ぜひカジュアルにお話ししましょう〜!🤝
pitta.me/matches/scbUsnxgXYJ…
Presteniko Septi R retweeted
WareTwin
Real-time 3D digital twin of an autonomous warehouse — 20 AMRs, deterministic simulation, explainable fleet scheduling, failure injection, VLM camera perception, AI Ops Copilot and what-if scenarios. React Three Fiber + FastAPI.
This article really made me re-think that the approaches are vastly different between AI personas and Digital Twin.
aiforinsightsleaders.substac…
AI personas help teams think.
Digital twins help teams decide.
28
Presteniko Septi R retweeted
🇯🇵 🚨 JAPAN: THREAT ACTOR CLAIMS 55 MILLION SOFTBANK CUSTOMER RECORDS FOR SALE
A threat actor on a dark web forum is offering a database they say holds 55 million records belonging to customers of SoftBank Corp., one of Japan's largest mobile carriers. The asking price is $500.
The actor claims the data includes:
* Names, email addresses, phone numbers, street addresses and dates of birth
* MSISDN, IMSI and ICCID (phone number and SIM identifiers), plan codes and device details
* Billing data such as payment methods and monthly bill amounts
⚠️ Analyst Note:
SIM identifiers paired with identity and billing data are ideal material for SIM-swap fraud and convincing carrier-themed phishing. A low price for a dataset this size can point to recycled or inflated data, and a similar SoftBank listing showing only column names circulated earlier this month. Unverified threat-actor claim, NOT confirmation that SoftBank was breached.
#DDW #DarkWeb #SoftBank #Japan #DataBreach #SIMSwap #ThreatIntelligence
How big the model scores in whatever benchmarks will not going to make all of your problems be solved.
14
Presteniko Septi R retweeted
Congrats to @nvidia for creating the best open-weight model on document parsing so far.
Nemotron 3.5 Parse takes the #1 spot on ParseBench, our comprehensive real-world benchmark for document understanding. The full list consists of 150+ frontier models, open-weight models, and specialized solutions.
It does this while remaining sub-1B parameters so it's also blazing fast ⚡️
Big thanks to @JensenHuang for the shoutout at GTC Berlin!
Nemotron release: huggingface.co/blog/nvidia/n…
ParseBench: parsebench.ai/
Presteniko Septi R retweeted
Researchers proved every LLM is secretly obsessed with Japan.
And they finally figured out why.
They published a paper investigating hidden cultural biases in large language models.
they tested 31,000 questions across 24 different languages on models like gemini, chatgpt, claude, and llama.
they stripped out all geographic hints and just asked the models to pick a place and answer.
the finding was universal.. across a wide range of models and languages, there is a pervasive "obsession" with japanese culture.
even when you prompt them in a language with zero ties to japan, they default to japanese norms, values, and aesthetics at a disproportionately high rate compared to its actual global demographic weight.
the craziest part? they tracked down exactly when the ai develops this obsession.. and it is not from reading the raw internet.
during pre-training, the models actually have a fairly balanced, global view of the world. the obsession gets injected during the safety phase.
when tech companies fine-tune these models to be "helpful, harmless, and polite," the ai starts mapping those corporate safety guidelines directly onto japanese cultural concepts. in the pursuit of creating the perfect, inoffensive, compliant assistant, they warped how the models view the world.
the quantization process (compressing models to run faster locally) actually amplifies this determinism. the smaller and faster your local model is, the more stereotypically it behaves, causing japanese cultural tropes to spike hard.
we spend all this time worrying about existential ai safety, but the real alignment problem is that our most advanced reasoning engines are hyper-fixated on japanese internet culture.
every model upgrade is reinforcing an invisible regional bias that dictates how they generate stories, facts, and code. the cultural commonsense of the internet is not universal..
it’s just localized to whatever the heaviest reddit and anime wiki users were obsessed with.
Presteniko Septi R retweeted
🔥 Introducing Koi 👻 🔥
A privacy-focused little ghost that lives on your Mac.
→ An AI that sees your screen, like HeyClicky
→ Dictation that types anywhere, like Wispr Flow
→ Call notes, like Granola
→ Documents, like Notion
→ Flashcards, like Anki
→ One calendar for every account, like Fantastical
One app instead of seven.
Your documents, calls and calendar stay on your Mac.
koiscribe.com
More below 👇
Presteniko Septi R retweeted
A strong Physical AI company needs some combination of:
1. Painkiller use case: expensive, dangerous, repetitive, dirty or labor-constrained work.
2. Clear ROI: customer payback measurable in months, not an abstract productivity story.
3. High utilization: a $100k robot working 2 hours/day has very different economics from one working 18.
4. Reliability: demo success ≠ production reliability.
5. Low intervention rate: constant teleoperation can destroy the economics.
6. Deployment velocity: how quickly can you go from signed customer → productive robot?
7. Uptime / MTBF: how often does something physically break?
8. Field service: who repairs it, how quickly, and at what cost?
9. Deployment repeatability: does every new customer require a custom engineering project?
10. Workflow integration: how much does the customer need to change around the robot?
11. Safety: what is the cost of failure in the real world?
12. Data flywheel: does every deployment make the system better?
13. Fleet learning: can what robot #100 learns improve robot #1,001?
14. Gross margin after reality: maintenance, installation, teleoperation, spare parts and on-site support all count.
15. Form-factor advantage: why does this task need this embodiment at all?
The biggest Physical AI companies won’t be built on the best demos. They’ll be built on the best real-world economics.






















